Elamur Research⌕
☰
home / papers / 2010 / Offline Arabic Handwriting Recognition Using Artificial Neural Network

Offline Arabic Handwriting Recognition Using Artificial Neural Network

Paper · 2010 · A.A Zaidan, B.B Zaidan, Hamid.A.Jalab et al.

Summary

This short paper looks at offline recognition of handwritten Arabic. Offline means the system sees only a scanned image of finished writing, with no record of the pen's movement. The authors explain why Arabic is hard. Letters are joined in a cursive way. A single letter can take up to four shapes depending on where it sits in a word. Letter heights vary, which makes noise harder to spot. Errors at the character level also carry over into word recognition.

The proposed system is a classic pipeline:

  1. Convert a selected image region to RGB.
  2. Binarize it.
  3. Segment the non-black pixels.
  4. Extract features.
  5. Pass the features to an artificial neural network for classification.

For training, characters are segmented by hand, then features are extracted and selected. The paper reports 99.62% recognition on a handwriting database. However, it does not name the database, describe the network architecture or features, or explain the train/test protocol. The work is useful mainly as an outline of a standard OCR pipeline and of the specific problems of Arabic script. It should not be relied on as a verified benchmark.

Key points

  • The task is offline handwriting recognition: only the final image is available. In online recognition, by contrast, pen coordinates over time are captured on a tablet.
  • The Arabic script properties listed are:
    • 28 letters.
    • Words run right to left, while numerals run left to right.
    • Most letters change shape at the start, middle or end of a word, or when standing alone, giving up to four forms.
    • Joinable letters are always connected.
    • Uneven letter heights make noise detection harder.
  • The pipeline is: region → RGB → binary image → segmentation of non-black pixels → feature extraction → ANN → recognized character.
  • The training stage uses hand-segmented characters, followed by feature extraction and feature selection.
  • The claimed accuracy is 99.62% with an ANN on an unnamed handwriting database. No baselines, error analysis or dataset statistics are given.

Interesting ideas

  • I1. The paper lists three criteria for good features: they should separate the classes, ignore irrelevant variation, and stay few enough to keep computation and training-data needs low. This is a compact checklist for judging any hand-designed feature set.
  • I2. The same symbol is treated as having several surface forms depending on its position in the word, so position becomes part of what the classifier must handle. This generalises to any unit whose appearance depends on where it occurs.
  • I3. Training runs on cleanly hand-segmented units, while testing runs on raw input through the full pipeline. This separates the classifier's quality from the quality of segmentation.
  • I4. The offline/online split shows that the same task can be much harder when only the final product is visible and the process trace is lost.
  • I5. The paper's single headline figure has no dataset name, split or baseline, which makes it impossible to verify. That gap is itself a lesson for evaluation design.

For Mova

How these ideas could serve Mova, our small open language model — its files, stages and tests:

  • I1 (tokens): Use these three criteria to audit the tagger's and morphology's feature templates, pruning features that only capture spelling noise.
  • I2 (tokens): Make position-sensitive surface variants (sentence-initial capitals, clitic and contraction forms) explicit features or normalisations before tagging, rather than letting them inflate the vocabulary.
  • I3 (tree): Evaluate parsers both on gold tokenization and on Mova's own tokenizer output, so tree errors are not blamed on the parser when the tokenizer caused them.
  • I4 (world): Wherever possible, keep the derivation trace (which rule or witness produced each fact) alongside the final world state, since an explainable pipeline loses information if it stores only end results.
  • I5 (learn): Every reported score should name the data, the split and a negative control, such as a trivial baseline, so that results like this are never accepted without checking.

Our summary of the paper, in our own words — not its text. The work belongs to its authors; read it at the links above.

 Fields, topics and concepts

📚 Shelves 5The field of science this work belongs to (OpenAlex fields).Computer Science · 16,667 worksComputer Vision and Pattern Recognition · 2,869 worksEngineering · 133 worksHuman-Computer Interaction · 141 worksMedia Technology · 34 works

🧭 Topics 1Research subjects and languages the work is about.Semitic languages · 2,908 works

Cite

Cite as A.A Zaidan, B.B Zaidan, Hamid.A.Jalab, Hamdan.O.Alanazi and Rami Alnaqeib. 2010. Offline Arabic Handwriting Recognition Using Artificial Neural Network. arXiv:1006.2809 · Journal of Computer Science and Engineering, Volume 1, Issue 1, p55-58, May 2010.

The paper itself, not this page. BibTeX from the library's record of the work. DOI 10.48550/arxiv.1006.2809 · arXiv 1006.2809

@misc{zaidan2010offline,
  title = {{Offline Arabic Handwriting Recognition Using Artificial Neural Network}},
  author = {A.A Zaidan, B.B Zaidan, Hamid.A.Jalab, Hamdan.O.Alanazi and Rami Alnaqeib},
  year = {2010},
  doi = {10.48550/arxiv.1006.2809},
  eprint = {1006.2809},
  archivePrefix = {arXiv},
  url = {https://doi.org/10.48550/arxiv.1006.2809}
}

References

16 references, 0 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. . F. Hussain and J. Cowell, “Character Recognition of Arabic and Latin Scripts”, Proceedings, IEEE International Conference on Information Visualization, pp. 51–56, 2000 find
  2. . K. Khatatneh ,"Probabilistic Artificial Neural Network For Recognizing the Arabic. Hand Written Characters", Journal of Computer Science 3 (12), 881-886, 2006 ↗
  3. . L.Hammami and D.Berkani ,"Recognition system for printed multi-font and multi-size Arabic characters", The Arabian Journal for Science and Engineering, Volume 27, Number 1B, pp:57-72, April, 2002 find
  4. . C. Papageorgiou and T. Poggio, “A trainable system for object detection,” International Journal of Computer Vision, vol. 38, no. 1, pp. 15–33, 2000. ↗
  5. . G. Guodong, S. Li, and C. Kapluk, “character recognition by support vector machines,” in Proc. IEEE International Conference on Automatic Face and Gesture Recognition, 2000, pp. 196–201. find
  6. . K. Jonsson, J. Matas, J. Kittler, and Y. Li, “Learning Support Vectors for face verification and recognition,” in Proc. IEEE Int Conf on AutomaticFace and Gesture Recognition, 2000. find
  7. . B. Heisele, P. Ho, and T. Poggio, “Face recognition with support vector machines: global versus component-based approach,” in ICCV, 2001, pp. 688– 694. ↗
  8. . A. Mohan, C. Papageorgiou, and T. Poggio, “Example-based object detection in images by Components,” IEEE Trans. Patt. Anal. Mach. Intell., vol. 23, pp. 349–361, 2001. ↗
  9. . B. Scholkopf, P. Simard, A. Smola, and V. Vapnik, “Prior knowledge in support vector kernels,” in Advances in Neural Inf. Proc. Systems, vol. 10. MIT Press, 1998, pp. 640–646. find
  10. . O. Chapelle, P. Haffner, and V. Vapnik, “SVMs for histogram-based image classification,” IEEE Transactions on Neural Networks, special issue on Support Vectors, 1999. find
  11. . F. Jing, M. Li, H. Zhang, and B. Zhang, “Support Vector Machines for region-based image retrieval,” in Proc. IEEE International Conference on Multimedia and Expo, 2003. find
  12. . S. Belongie, C. Fowlkes, F. Chung, and J. Malik, “Spectral partitioning with indefinite kernels using the nystr¨om extention,” in ECCV, part III, Copenhagen, Denmark, may 2002, find
  13. . C. Wallraven, B. Caputo, and A. Graf, “Recognition with local features: the kernel recipe,” in Proceedings of the International Conference on Computer Vision, vol. I, 2003, p. 257ff. find
  14. . T. Evgeniou, M. Pontil, and T. Poggio, “Regularization networks and support vector machines,” Advances in Computational Mathematics, vol. 13, pp. 1–50, 2000. ↗
  15. . V. Vapnik, The nature of statistical learning Theory. John Wiley and sons, New York, 1995. find
  16. . D. M. J. Tax and R. Duin, “Uniform object generation for optimizing one-class classifiers,” Journal of Machine Learning Research, Special Issue on Kernel methods, no. 2, pp. 155–173, 2002 Aos Alaa Zaidan: He obtained his 1st Class Bachelor degree in Computer Engineering from university of Technology / Baghdad followed by master in data communication and computer network from University of Malaya. He led or member for many funded research projects and He has published more than 60 papers at various international and national conferences and journals, His interest area are Information security (Steganography and Digital watermarking), Network Security (Encryption Methods) , Image Processing (Skin Detector), Pattern Recognition , Machine Learning (Neural Network, Fuzzy Logic and Bayesian) Methods and Text Mining and Video Mining. .Currently, he is PhD Candidate on the Department of Electrical & Computer Engineering / Faculty of Engineering / Multimedia University / Cyberjaya, Malaysia. He is members IAENG, CSTA, WASET, and IACSIT. He is reviewer in the (IJSIS, IJCSN, IJCSE and IJCIIS). Bilal Bahaa Zaidan: He obtained his bachelor degree in Mathematics and Computer Application from Saddam University/Baghdad followed by master in data communication and computer network from University of Malaya. He led or member for many funded research projects and He has published more than 60 papers at various international and national conferences and journals, His interest area are Information security (Steganography and Digital watermarking), Network Security (Encryption Methods) , Image Processing (Skin Detector), Pattern Recognition , Machine Learning (Neural Network, Fuzzy Logic and Bayesian) Methods and Text Mining and Video Mining. .Currently, he is PhD Candidate on the Department of Electrical & Computer Engineering / Faculty of Engineering / Multimedia University / Cyberjaya, Malaysia. He is members IAENG, CSTA, WASET, and IACSIT. He is reviewer in the (IJSIS, IJCSN, IJCSE and IJCIIS). Dr.Hamid.A.Jalab: Received his B.Sc degree from University of Technology, Baghdad, Iraq. MSc & Ph.D degrees from Odessa Polytechnic National State University 1987 and 1991, respectively. Presently, Visiting Senior Lecturer of Computer System and Technology, Faculty of Computer Science and Information Technology, University of Malaya, Malaysia. various international and national conferences and journals, His interest area are Information security (Steganography and Digital watermarking), Network Security (Encryption Methods) , Image Processing (Skin Detector), Pattern Recognition , Machine Learning (Neural Network, Fuzzy Logic and Bayesian) Methods and Text Mining and Video Mining. © 2010 JCSE http://sites.google.com/site/jcseuk/Hamdan Al-Anazi: has obtained his bachelor dgree from “King Suad University”, Riyadh, Saudi Arabia. He worked as a lecturer at Health College in the Ministry of Health in Saudi Arabia, then he worked as a lecturer at King Saud University in the computer department. Currently he is Master candidate at faculty of Computer Science & Information Technology at University of Malaya in Kuala Lumpur, Malaysia. His research interest on Information Security, cryptography, steganography and digital watermarking, He has contributed to many papers some of them still under reviewer. Rami Alnaqeib - he is master student in the Department of Information Technology / Faculty of Computer Science and Information Technology/University of Malaya / Kuala Lumpur/Malaysia, He has contribution for many papers at international conferences and journals. ↗

Updated

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/zaidan-2010-offline-arabic-handwriting-recognition-using/